Electro-elastic material modeling with neural networks

Congratulations to our Ph.D. student Dominik Klein, whose latest article “Finite electro-elasticity with physics-augmented neural networks” just appeared in the journal Computer Methods in Applied Mechanics and Engineering! In a great collaboration with Rogelio Ortigosa & Jesús Martínez-Frutos from TU Cartagena in Spain, we extended our physics-augmented neural network constitutive models to electro-elastic material behavior at finite deformations.

Using coupled invariants and convex neural networks, the approach is thermodynamically consistent, fulfills all physical and mathematical constitutive requirements, and attains high accuracy even when only small amounts of calibration data are available. Furthermore, could demonstrate its applicability on an analytically homogenized, transversely isotropic rank-one laminated composite and a numerically homogenized cubic metamaterial.

The article is available for download at https://doi.org/10.1016/j.cma.2022.115501 (50 days free access) and https://arxiv.org/abs/2206.05139.